sglang 0.4.2__py3-none-any.whl → 0.4.2.post1__py3-none-any.whl

sglang/srt/models/qwen2_vl.py

@@ -30,12 +30,10 @@ import numpy as np
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from einops import rearrange, repeat
+from einops import rearrange
 from vllm.model_executor.layers.activation import QuickGELU
 
 from sglang.srt.configs import Qwen2VLConfig, Qwen2VLVisionConfig
-from sglang.srt.distributed import parallel_state
-from sglang.srt.distributed import utils as dist_utils
 from sglang.srt.hf_transformers_utils import get_processor
 from sglang.srt.layers.attention.vision import VisionAttention
 from sglang.srt.layers.linear import ColumnParallelLinear, RowParallelLinear
@@ -118,6 +116,7 @@ class Qwen2VisionBlock(nn.Module):
         mlp_ratio: float,
         act_layer: Type[nn.Module] = QuickGELU,
         norm_layer: Type[nn.Module] = None,
+        attn_implementation: Optional[str] = "sdpa",
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
@@ -126,12 +125,24 @@ class Qwen2VisionBlock(nn.Module):
         self.norm1 = norm_layer(dim)
         self.norm2 = norm_layer(dim)
         mlp_hidden_dim = int(dim * mlp_ratio)
+        if attn_implementation == "sdpa":
+            use_context_forward = False
+            use_full_precision_softmax = False
+        elif attn_implementation == "flash_attention_2":
+            use_full_precision_softmax = False
+            use_context_forward = True
+        elif attn_implementation == "eager":
+            use_full_precision_softmax = True
+            use_context_forward = False
 
         self.attn = VisionAttention(
             embed_dim=dim,
             num_heads=num_heads,
             projection_size=dim,
             use_qkv_parallel=False,
+            use_context_forward=use_context_forward,
+            use_full_precision_softmax=use_full_precision_softmax,
+            flatten_batch=True,
             quant_config=quant_config,
         )
         self.mlp = Qwen2VisionMLP(
@@ -286,7 +297,6 @@ class Qwen2VisionTransformer(nn.Module):
         norm_layer = partial(nn.LayerNorm, eps=norm_eps)
         head_dim = embed_dim // num_heads
         self.rotary_pos_emb = Qwen2VisionRotaryEmbedding(head_dim // 2)
-
         self.blocks = nn.ModuleList(
             [
                 Qwen2VisionBlock(
@@ -294,6 +304,7 @@ class Qwen2VisionTransformer(nn.Module):
                     num_heads=num_heads,
                     mlp_ratio=mlp_ratio,
                     norm_layer=norm_layer,
+                    attn_implementation="sdpa",
                     quant_config=quant_config,
                 )
                 for _ in range(depth)
@@ -482,10 +493,6 @@ class Qwen2VLForConditionalGeneration(nn.Module):
                 opensource models), the shape will be `(3, seq_len)`,
                 otherwise it will be `(seq_len,).
                 (Use input_metadata.mrope_positions to replace it)
-            pixel_values: Pixel values to be fed to a model.
-                `None` if no images are passed.
-            image_grid_thw: Tensor `(n_images, 3)` of image 3D grid in LLM.
-                `None` if no images are passed.
         """
         if getattr(self.config, "rope_scaling", {}).get("type", None) == "mrope":
             positions = forward_batch.mrope_positions
@@ -540,15 +547,18 @@ class Qwen2VLForConditionalGeneration(nn.Module):
                     num_image_tokens = self.calculate_num_image_tokens(
                         image_grid_thws[idx]
                     )
+
                     left_idx = start_idx + (image_offset - prefix_len)
                     right_idx = (
                         start_idx + (image_offset - prefix_len) + num_image_tokens
                     )
+
                     inputs_embeds[left_idx:right_idx] = image_embeds[
                         image_embeds_offset : image_embeds_offset + num_image_tokens
                     ]
                     image_embeds_offset += num_image_tokens
 
+        input_ids = None
         hidden_states = self.model(
             input_ids=input_ids,
             positions=positions,

sglang/srt/server_args.py

@@ -163,6 +163,7 @@ class ServerArgs:
     # Custom logit processor
     enable_custom_logit_processor: bool = False
     tool_call_parser: str = None
+    enable_hierarchical_cache: bool = False
 
     def __post_init__(self):
         # Set missing default values
@@ -892,6 +893,11 @@ class ServerArgs:
             default=ServerArgs.tool_call_parser,
             help="Specify the parser for handling tool-call interactions. Options include: 'qwen25', 'mistral', and 'llama3'.",
         )
+        parser.add_argument(
+            "--enable-hierarchical-cache",
+            action="store_true",
+            help="Enable hierarchical cache",
+        )
 
     @classmethod
     def from_cli_args(cls, args: argparse.Namespace):

sglang/srt/utils.py

@@ -444,8 +444,6 @@ def load_image(image_file: Union[str, bytes]):
     else:
         raise ValueError(f"Invalid image: {image}")
 
-    # if image_size is None:
-    #     image_size = image.size
     return image, image_size
 
 

sglang/utils.py

@@ -373,3 +373,45 @@ class TypeBasedDispatcher:
             if isinstance(obj, ty):
                 return fn(obj)
         raise ValueError(f"Invalid object: {obj}")
+
+
+def trim_overlap(existing_text, new_chunk):
+    """
+    Finds the largest suffix of 'existing_text' that is a prefix of 'new_chunk'
+    and removes that overlap from the start of 'new_chunk'.
+    """
+    max_overlap = 0
+    max_possible = min(len(existing_text), len(new_chunk))
+    for i in range(max_possible, 0, -1):
+        if existing_text.endswith(new_chunk[:i]):
+            max_overlap = i
+            break
+    return new_chunk[max_overlap:]
+
+
+def stream_and_merge(llm, prompt, sampling_params):
+    """
+    1) Streams the text,
+    2) Removes chunk overlaps,
+    3) Returns the merged text.
+    """
+    final_text = ""
+    for chunk in llm.generate(prompt, sampling_params, stream=True):
+        chunk_text = chunk["text"]
+        cleaned_chunk = trim_overlap(final_text, chunk_text)
+        final_text += cleaned_chunk
+    return final_text
+
+
+async def async_stream_and_merge(llm, prompt, sampling_params):
+    """
+    Streams tokens asynchronously, removes chunk overlaps,
+    and yields the cleaned chunk in real time for printing.
+    """
+    final_text = ""
+    generator = await llm.async_generate(prompt, sampling_params, stream=True)
+    async for chunk in generator:
+        chunk_text = chunk["text"]
+        cleaned_chunk = trim_overlap(final_text, chunk_text)
+        final_text += cleaned_chunk
+        yield cleaned_chunk  # yield the non-overlapping portion

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.4.2"
+__version__ = "0.4.2.post1"

{sglang-0.4.2.dist-info → sglang-0.4.2.post1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: sglang
-Version: 0.4.2
+Version: 0.4.2.post1
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -333,7 +333,7 @@ Requires-Dist: sglang[test]; extra == "dev-cpu"
 | [**Slides**](https://github.com/sgl-project/sgl-learning-materials?tab=readme-ov-file#slides) |
 
 ## News
-- [2025/01] 🔥 SGLang provides day one support for DeepSeek V3/R1 models on NVIDIA and AMD GPUs with DeekSeek-specific optimizations. ([instructions](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3), [AMD blog](https://www.amd.com/en/developer/resources/technical-articles/amd-instinct-gpus-power-deepseek-v3-revolutionizing-ai-development-with-sglang.html))
+- [2025/01] 🔥 SGLang provides day one support for DeepSeek V3/R1 models on NVIDIA and AMD GPUs with DeepSeek-specific optimizations. ([instructions](https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3), [AMD blog](https://www.amd.com/en/developer/resources/technical-articles/amd-instinct-gpus-power-deepseek-v3-revolutionizing-ai-development-with-sglang.html))
 - [2024/12] 🔥 v0.4 Release: Zero-Overhead Batch Scheduler, Cache-Aware Load Balancer, Faster Structured Outputs ([blog](https://lmsys.org/blog/2024-12-04-sglang-v0-4/)).
 - [2024/09] v0.3 Release: 7x Faster DeepSeek MLA, 1.5x Faster torch.compile, Multi-Image/Video LLaVA-OneVision ([blog](https://lmsys.org/blog/2024-09-04-sglang-v0-3/)).
 - [2024/07] v0.2 Release: Faster Llama3 Serving with SGLang Runtime (vs. TensorRT-LLM, vLLM) ([blog](https://lmsys.org/blog/2024-07-25-sglang-llama3/)).
@@ -372,7 +372,7 @@ Learn more in the release blogs: [v0.2 blog](https://lmsys.org/blog/2024-07-25-s
 [Development Roadmap (2024 Q4)](https://github.com/sgl-project/sglang/issues/1487)
 
 ## Adoption and Sponsorship
-The project is supported by (alphabetically): AMD, Baseten, Cursor, DataCrunch, Etched, Hyperbolic, Jam & Tea Studios, LinkedIn, LMSYS.org, Meituan, NVIDIA, RunPod, Stanford, UC Berkeley, UCLA, xAI, 01.AI.
+The project is supported by (alphabetically): AMD, Baseten, Cursor, DataCrunch, Etched, Hyperbolic, Jam & Tea Studios, LinkedIn, LMSYS.org, Meituan, Novita AI, NVIDIA, RunPod, Stanford, UC Berkeley, UCLA, xAI, 01.AI.
 
 ## Acknowledgment and Citation
 We learned the design and reused code from the following projects: [Guidance](https://github.com/guidance-ai/guidance), [vLLM](https://github.com/vllm-project/vllm), [LightLLM](https://github.com/ModelTC/lightllm), [FlashInfer](https://github.com/flashinfer-ai/flashinfer), [Outlines](https://github.com/outlines-dev/outlines), and [LMQL](https://github.com/eth-sri/lmql). Please cite the paper, [SGLang: Efficient Execution of Structured Language Model Programs](https://arxiv.org/abs/2312.07104), if you find the project useful.

{sglang-0.4.2.dist-info → sglang-0.4.2.post1.dist-info}/RECORD

@@ -9,8 +9,8 @@ sglang/check_env.py,sha256=4OqpZaEJOfBM6-vtPILto5kqDmgiZM1Koc7lK78A7CI,8427
 sglang/global_config.py,sha256=fnT0U9vlHdGaQFKN9tYTnUF4-eVW4HYQURd5zvPtrg0,1286
 sglang/launch_server.py,sha256=mDXfwha8LHpWQJekcCosR98QhCQsbmilsBlI5jAIgg0,420
 sglang/llama3_eval.py,sha256=gWSboDchIGybIce88bJlrCG0yiLZ513mw4gcutJlzGM,10017
-sglang/utils.py,sha256=wvLVVC8U2qIhCSCrtzvV3wXapvJweir1XDNdpfoPFRM,11934
-sglang/version.py,sha256=6hfVa12Q-nXyUEXr6SyKpqPEDJW6vlRHyPxlA27PfTs,22
+sglang/utils.py,sha256=7HpOrPBhMivWH719m7Dy1rjrAXOAsnqelpwNBBbvjqs,13319
+sglang/version.py,sha256=BObAQyMJTgNEQbPpM5x4R8aeAPCZ_eHVSXPwL90NUlk,28
 sglang/lang/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/lang/chat_template.py,sha256=v4SyYViPHX3i3XT46F7vlARn4UaSiP3PBpTGtzO6uRY,17006
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
@@ -33,9 +33,9 @@ sglang/srt/hf_transformers_utils.py,sha256=_24uqCkZ4dvS9Uc5p2cCzX0Q8ShUzrh_Hp6mv
 sglang/srt/mm_utils.py,sha256=1ScBunw_x4W8ebM_AcJ62-1T2mfT8NlMJqdAhkF1lb0,12367
 sglang/srt/model_parallel.py,sha256=eLXZhvJ4wG6dh0FontNCIdVZvHYdWgaeY-5cu7TD9tE,6078
 sglang/srt/server.py,sha256=PrQb9r6L9syWHKlggbbiQYsKtpwSmECqozRbf8qnoV8,874
-sglang/srt/server_args.py,sha256=GN9NAKDkSWpMQWoTBzHyvp-UimfKKpwAmzgiwUJTe4A,39792
+sglang/srt/server_args.py,sha256=opURYsAG9anR5EINNq45f8GJv3NLDllhP9AlwpJ3lK8,39997
 sglang/srt/torch_memory_saver_adapter.py,sha256=--FgbrcvJxTcRe856plD9ktqgrHGPTE18eZCJlE50hY,1255
-sglang/srt/utils.py,sha256=jzHsVQDYF25Z7rPjiaO4w5iz7ZskRRZxTvEiUeFcSJw,46380
+sglang/srt/utils.py,sha256=yIQ5XtfJa_jPDKTzxqXnCdbhA2kKMihzcP4fSAWU4bs,46317
 sglang/srt/configs/__init__.py,sha256=Nvwtif0X9IYUtj0aL9XvAo_RRZcxTshsaliwc8djooU,347
 sglang/srt/configs/chatglm.py,sha256=j-b0YkdYUmQm2y1kNmMJtKeACxWKmBbvNNkDWbs6kbI,2907
 sglang/srt/configs/dbrx.py,sha256=tdhIkXAQl1yr0MxqFmsDG1E0e2puRTTKm6UTyANBLac,11005
@@ -71,8 +71,8 @@ sglang/srt/layers/logits_processor.py,sha256=_3TZNUbvjmw63ywBv6V6WU87G1TErMaXGa7
 sglang/srt/layers/parameter.py,sha256=sX6aB69qbD6jRqQeOfXqK_ueyyZlXCeC0AlglbsRPcM,14901
 sglang/srt/layers/pooler.py,sha256=rj2lygvleBnyLCBZ8I11HGMgpfIDsT0l3PIkshJwdu4,1606
 sglang/srt/layers/radix_attention.py,sha256=tPjJA3P9kuFBk2QWFTgOI8UbVUFLVDZgFaQWuokx894,2234
-sglang/srt/layers/rotary_embedding.py,sha256=CdnkPxUtef-o29i1G2p1an3H7sEgScWLvI_XVoMaPbo,43444
-sglang/srt/layers/sampler.py,sha256=T_Lvjc7PhmOUhNAeSoI14DG9EO4XFve0z3wgWYy7YLU,9769
+sglang/srt/layers/rotary_embedding.py,sha256=tEvy-IAi7GaI5PYFwV30Rek3m6oUHKYuGSa23FXWXSE,44100
+sglang/srt/layers/sampler.py,sha256=4GFSuaNrvnM9S-GPtqTeiWu3I0XZvxM7B48NdSLwpW4,9934
 sglang/srt/layers/torchao_utils.py,sha256=Ws24FdRBSkTpyeyA6bQrdDm-W5wfDxKvSIPUSahyMfA,4063
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=txcjkuSDa6gZwESKj8X-HSLhAnMmDXL0FmFWY9SKqik,22155
 sglang/srt/layers/attention/__init__.py,sha256=KlQ0fl-o9v_NxBDhNZ4dPW2uQ2HeJjLm-0MTMWgaa28,2980
@@ -80,11 +80,11 @@ sglang/srt/layers/attention/double_sparsity_backend.py,sha256=QEDF8tQKMkh-nbt4jH
 sglang/srt/layers/attention/flashinfer_backend.py,sha256=XUyR97-WSyE6esq4r4XOcvXRtEJm8JOZ6MrXE-YfsYM,33949
 sglang/srt/layers/attention/torch_native_backend.py,sha256=KrcAqTLVZLtwgOmB0xhwUUsX32M-5LYZpNxaRNT4VuA,9252
 sglang/srt/layers/attention/triton_backend.py,sha256=P329qd6i7XfgB2UH7KXNid67v-kziV1sgcAuh3RWna8,6654
-sglang/srt/layers/attention/vision.py,sha256=mn8fruFob-Cif0_6V5P6W-2lCqJOiIsvsmYtH-bEvcU,6643
+sglang/srt/layers/attention/vision.py,sha256=zLjKmzUlkgq1RFcP3b4EPArOAKovoaDLgYfM5SyB2wM,13181
 sglang/srt/layers/attention/triton_ops/decode_attention.py,sha256=ltWcZ00ugpglSYvszpGb-UCpGIixdG25cWtSrOOOMik,17943
 sglang/srt/layers/attention/triton_ops/double_sparsity_attention.py,sha256=1pSXfY3EEaM7iRN_uElHnAfsrJMhTFbu9fj8Z0O2PbE,21480
 sglang/srt/layers/attention/triton_ops/extend_attention.py,sha256=DWOZXSTVN5ZbcFjDjcqs-nPdUkxSwum0SVXhVKqwh2g,11688
-sglang/srt/layers/attention/triton_ops/prefill_attention.py,sha256=lojFXRZMLWkzS2Y8uxaolnQhXaWKG19mCAWaF5KQeiI,6087
+sglang/srt/layers/attention/triton_ops/prefill_attention.py,sha256=Y66gZ37u0GKMPtI8n5MbO6uOxRuGEmKIG0IPbJTOqAM,6213
 sglang/srt/layers/moe/fused_moe_native.py,sha256=OEWpM93X5tJG4-rwz5qmdpTzEUR73zun29YRV3bZglY,4269
 sglang/srt/layers/moe/topk.py,sha256=qcWDUVvEV6TIO_idymStylkpPp6dMk-wbYj2Zq4ZYJ0,7057
 sglang/srt/layers/moe/ep_moe/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -187,7 +187,7 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=bofJhiDnRNqD2D20QV7CPNf2S
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json",sha256=Uz5X80VcNBOaxshwVNUEittHk2zqB4HQCfTJ4TPG5aM,3274
 sglang/srt/layers/quantization/__init__.py,sha256=_Sba1KQnmZNKGDKM1MfBs2T3uDqOHfeW6IHO2mTUvfs,4471
 sglang/srt/layers/quantization/base_config.py,sha256=daK9p0aijMszLUm1W4Pc33FK87MdqYK1NoWFKif-j80,4599
-sglang/srt/layers/quantization/fp8.py,sha256=XjLGHFPgX0NBuXa8eOglZ8TPMvXMNges0l4gDdcumRE,34866
+sglang/srt/layers/quantization/fp8.py,sha256=ibttPVCUsCQ0LXy7FUb8wnzqGcGZQXQLqwCB4a2fai4,35160
 sglang/srt/layers/quantization/fp8_kernel.py,sha256=cYF4ckqrUyhCO9Ha7zi05R8EhRaqSa8rFpYisz-9Ed0,10743
 sglang/srt/layers/quantization/fp8_utils.py,sha256=7v-RNwuYXa-gPO3msRDB0Z3uajOQMYd2Cj0NMoq1hg4,4148
 sglang/srt/layers/quantization/int8_kernel.py,sha256=t_BLVf8XjOyn7S3Lu3B4hXvw8DvTg4Anco7TNadL58U,1436
@@ -247,21 +247,21 @@ sglang/srt/managers/cache_controller.py,sha256=DXnIunJgtTws1WF2vZOYVQe56vacV7Mn4
 sglang/srt/managers/configure_logging.py,sha256=aY9xExurz7t_IdItd-9GuVuM7kEGB8_bRryhZxKdu9o,1542
 sglang/srt/managers/data_parallel_controller.py,sha256=b64aC6iLr5RolJyNQnT-yTQ_TSI9DDLtuABf_TPTUrM,9421
 sglang/srt/managers/detokenizer_manager.py,sha256=A-tZi9VPkrIAVteQItYUY-07V1rWmySFHNcVf8qAdPI,9578
-sglang/srt/managers/image_processor.py,sha256=dEjEWzrmJyEXhr5sKBw4BEUEjla8CNdkzFGfogPGmFY,19103
+sglang/srt/managers/image_processor.py,sha256=s1QH9cSzT_nnitc6idzFjuGDp-pDnMTpbVZoQfzdSXU,20671
 sglang/srt/managers/io_struct.py,sha256=1Z6MCVI1LN2lS_7e8WHkpVNT_LW62mE-jpZ2Jn_FAtE,18267
 sglang/srt/managers/schedule_batch.py,sha256=oP6ygJUOmo6PuXcA_wecRvOOa_WdpwmIyCPSgJy4qAc,48743
 sglang/srt/managers/schedule_policy.py,sha256=Qero_lwPEb7bM87qjWtYijGyRhtY0mMwjWP6SbjvaUE,18260
-sglang/srt/managers/scheduler.py,sha256=QGGSau-ydbRzIFdCvE63Na-tpYKHJj_QL1d9raogvXc,70019
+sglang/srt/managers/scheduler.py,sha256=akwBfBcNgpCXY1vp3FlD5-bOUMKfUBR5AC3XzSBRDYQ,70757
 sglang/srt/managers/session_controller.py,sha256=WXRbtninVEVM0rQYiXFzOwsDph0TNj1L2sRCWQF0dSg,5571
 sglang/srt/managers/tokenizer_manager.py,sha256=TjhX0IeFCmk31PDmtVV7Ilc8rqI361XUf_p2KO3ai7s,38669
 sglang/srt/managers/tp_worker.py,sha256=OiHpFR9Hy1GpgLEkTDsykBiFuv1VKmkjQS58gQVPQIs,8126
 sglang/srt/managers/tp_worker_overlap_thread.py,sha256=7p6zREndc4a9fmYfqW4iY9IYANxdoAioaf0hU92-8Ow,8893
 sglang/srt/managers/utils.py,sha256=5i75uLlQOF_5CaT02CrWtwozMTtwTg2_nLP8Dtr-JZQ,1536
-sglang/srt/mem_cache/base_prefix_cache.py,sha256=QC8HS8RC5DXu14kyXsxAgEUsn0f932p2DjqzbKjc6Bs,962
-sglang/srt/mem_cache/chunk_cache.py,sha256=R2gHAuqKd5ayQW3NnsgoGUH31---Z5izCDyCqLL0FjQ,2524
+sglang/srt/mem_cache/base_prefix_cache.py,sha256=qxgpSHm3qtMdab4U35Mr2BE9TQNjElrnrNMTwL_Osdo,1049
+sglang/srt/mem_cache/chunk_cache.py,sha256=hc_reKKvoI4r8xkgf4I4eIkwXWTJC2ZXaQWuODQZnx0,2572
 sglang/srt/mem_cache/flush_cache.py,sha256=GYcxmNXh4hsMpFfNOuCTpKilW7guZwTtAg_usVeM3J0,979
 sglang/srt/mem_cache/memory_pool.py,sha256=9ud97u1cXnN6O0qlR8tv8woN_20gqisTV6aBgHqhinc,19682
-sglang/srt/mem_cache/radix_cache.py,sha256=c5voySV5L855c0G9cBEc9iQ4nR7PDDmg0V6fWWJHcq4,10945
+sglang/srt/mem_cache/radix_cache.py,sha256=hVILXvc5PauHuLTeyZbm3NCf3AOimaAuXjll53MSLeU,11754
 sglang/srt/metrics/collector.py,sha256=_yl0_paSARxS1ypZgd-pLJ29tMizolHuwROX21dOXTk,7326
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
 sglang/srt/model_executor/cuda_graph_runner.py,sha256=3_s7zmLn9d0pVCxZd43bYtPmgkbe1kcRaNZbryMMjPU,18520
@@ -296,20 +296,20 @@ sglang/srt/models/llava.py,sha256=xrkg8sht8tBOID7427IEZtHL-KKWfEivDe2NqGjTSAs,26
 sglang/srt/models/llavavid.py,sha256=dYUkKfHoE15vF_VXA_s_ICCTUMSmSgvP181fk8dUi0g,12185
 sglang/srt/models/minicpm.py,sha256=hVWri0-3sAiuGOMcIhGL2GphQZ13qBcLXuLTsQVALGY,13720
 sglang/srt/models/minicpm3.py,sha256=DZ7LltHsyDq8iE7nMi5C9gLzYcQrAIZYkRmx6lCuAgo,24683
-sglang/srt/models/minicpmv.py,sha256=lgWqj1bWMDvPHPE5POVEjhnY-_qMSidkbsBLMYBtDgM,43181
+sglang/srt/models/minicpmv.py,sha256=AHLQkg2Klimkr7-M3vOT0y5OeFBR6ftPkvDqzGs5hWM,45051
 sglang/srt/models/mistral.py,sha256=EYifJUUzN2Z2-iL37eJiNZF_DB0H4pa0mKlgYRIxM70,838
 sglang/srt/models/mixtral.py,sha256=ybArp6vx7VTrjQ3kqH1FHJ1gQzsFPI5vv1C-Pnix6ws,14520
 sglang/srt/models/mixtral_quant.py,sha256=_gy4gKwFX6BNlU6xE-n0N3vVNhftxgZjWEDKTCKV_2M,14019
-sglang/srt/models/mllama.py,sha256=vK80nRlSH7hFjBeqVRGVJ4XepfxkpY4HJPpvZ15CzMA,37751
+sglang/srt/models/mllama.py,sha256=q2JNkfqEZghmHWhcgX_YaaKVpXwTHnrz7z1TuzfHEjA,36340
 sglang/srt/models/olmo.py,sha256=-t5s3DI-CxiMqRAvKS73NTMNrRpQRD8eh2VabCNYDnE,11699
 sglang/srt/models/olmo2.py,sha256=Wg4mo53c3OIAWmAMZ-TR9VRzSfKqhBZixqvrF8AbIJg,13430
 sglang/srt/models/olmoe.py,sha256=luqgdyCYJTFyhaRfZElWSFV17ee6FjfU0CpemMmsTS8,15147
 sglang/srt/models/phi3_small.py,sha256=jVKH2twKfELtqyjMWjH8CnyXlCKEkYtiUUnx18k9OLQ,14799
 sglang/srt/models/qwen.py,sha256=dg_sVrh7I58Q_LevvO2d5dFZi1T19V2czNh8-9nPUaE,9901
-sglang/srt/models/qwen2.py,sha256=caVvsTu0QteR9Q65p81JEgESSV9_nP3yPaShRMu6uDY,14936
+sglang/srt/models/qwen2.py,sha256=igq-a61CQgH26xnim6c3yeWUCHiN_Nboxg4iu7oy7bo,15072
 sglang/srt/models/qwen2_eagle.py,sha256=KTtejEezdLfd_odg3Na1i5kBk7W-YFg9hImfWyrMgVc,4288
 sglang/srt/models/qwen2_moe.py,sha256=GWi5nuaQWifPmyC3ld2G1wZJS5Xva6-1yjCUrNcGhkY,16539
-sglang/srt/models/qwen2_vl.py,sha256=r0OmFH8OcsIZ96fKqXaAWGLUe6oTVW_w6Gt5PChYUXE,23139
+sglang/srt/models/qwen2_vl.py,sha256=M-8abTK2Id36Ba6TfvAQHwbdbDNcbgFAxNw6gXkbBCU,23475
 sglang/srt/models/registry.py,sha256=inKh9iwOp3LFYm3nqujg-OtABClOP-ifc1stA9cZegA,3434
 sglang/srt/models/stablelm.py,sha256=dO6EwFFiBWn-8yxV9tb3OtjNe9D0dF57Z298g7SmrhU,11308
 sglang/srt/models/torch_native_llama.py,sha256=X0AvlREIysazwFezqndRza7ZCWQ-R1hePoLW0brH4As,19131
@@ -347,8 +347,8 @@ sglang/test/test_layernorm.py,sha256=IacByD5d-stXjzBz8Ypamc7povlcedpKPbb_4JLgo3c
 sglang/test/test_programs.py,sha256=aUV9Ex_B714ph7ytv6W3J7sdGDKC6lGIhUy95Yg6AHQ,18878
 sglang/test/test_utils.py,sha256=BU6lAX3bu3TNQZqVC9UPnyq3I7iV5kigHQKJx7UNlOQ,26192
 sglang/test/srt/sampling/penaltylib/utils.py,sha256=CjxHgywh0hx_87iynzQt_ztHu6zBVuE-YrZ-XPmW6U4,12906
-sglang-0.4.2.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.2.dist-info/METADATA,sha256=tSq4P2Rhddw25uHR313Islv7DCbaPFSUrZzD8C8_pas,23224
-sglang-0.4.2.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-sglang-0.4.2.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.2.dist-info/RECORD,,
+sglang-0.4.2.post1.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.2.post1.dist-info/METADATA,sha256=DAQlsgAw08lMBLrkY88_9VfhV50Mt3Vi0H3uhM_RJOU,23241
+sglang-0.4.2.post1.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+sglang-0.4.2.post1.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.2.post1.dist-info/RECORD,,